Professional Documents
Culture Documents
AND
SCOTT F. RICHARD
Cumegie-Mellon Universit)
Motivated by recent work of Huang and Kreps it is shown that the existence of
a general equilibrium in an exchange economy whose commodity space is a vector
lattice is guaranteed if (in addition to standard hypotheses) the price space is also
a lattice. This is very weak in contrast with the usual requirement that the topology
of the commodity space be locally solid (i.e., continuity of the lattice operations),
a condition violated in models of Jones and of Huang and Kreps. The key to our
proof is a disaggregated approach to the construction of supporting prices. Journal
of Economic Literature Classification Numbers: 021,022. F: 1991 Academic Press. Inc.
1. INTRODUCTION
* Our debt to C. F. Huang and D. Kreps will be obvious. The problem treated in the paper
was proposed to us by them. R. Aliprantis, C. F. Huang, and W. Zame gave helpful and
patient answers to our many consulations. The paper was presented at the NBER General
Equilibrium Conference held at Stanford, March 1987. Thanks are due to the audience and
also to R. Anderson and G. Morin for a very careful reading.
+ NSF support is gratefully acknowledged.
1
0022-0531/91 $3.00
Copyright 8c-i 1991 by Academic Press, Inc.
All rights of reproduction m any form reserved
2 MAS-COLELL AND RICHARD
3. DISCUSSION
generalizes the results that rely on locally solid topologies which, in turn,
constitute a generalization of the seminal work of Bewley [7] for the
space L, .
To be specific we have required our compactness hypothesis to be the
a(L, L*)-compactness of X. But as in Mas-Cole11 [ 141 a weaker condition
suffices, namely the compactness of the utility set 0 = { (ur , .... uN): 0 < U, <
ui(xi) for some x E X}. Also, the careful reader will realize that without a
word of the proof being changed, the theorem remains valid if “o-uniform
proper” is interpreted as “o-uniform proper on [0, 01.”
The most interesting applications of our theorem are, of course to exam-
ples not covered by previous theory. Generally speaking, those are likely to
occur whenever L has a weak topology or when L, is small.
The commodity differentiation model studied by Jones [lo] corresponds
to L = ca(K), K compact, L, = (2 EL: ,-(A) 2 0 for every Bore1 set A c K},
and the weak-star topology on L. Here the topology is too weak for the
lattice operations to be continuous. (For K= [0, 1] consider hllk - 6, as
k + 0.) Nonetheless, Jones’ model is included in ours since L* = C(K), with
the induced positive orthant, is a vector lattice (more precisely, a sublattice
of the order dual). Also, our uniform properness condition (which has a
Lipschitz-like character) is somewhat weaker than his smoothness-like
hypothesis. For example, with K= [0, l] the utility function U(X) =
min{ .u( [0, 1I), 3 + x( [0, 11) - i j t tix(t)} is admissible by us but not by
him.
The family of intertemporal preferences models of Huang and Kreps
provides another example. The simplest specimen of the family is
L = (right continuous functions of bounded variation on [0, l] } with the
II.11I norm and positive cone L, = {x E L : x is nonnegative and non-
decreasing}. Equivalently, L = ca( [0, 11) with the standard positive cone
and the topology induced by the duality with C,(K), the Lipschitz con-
tinuous functions on [0, 11. Here the positive orthant L, is too small for
the lattice operations to be continuous. However, for this space the dual is
L,([O, 11) (equivalently L* = C,([O, 11)) which with the induced order
can be shown to be a vector lattice. (This is obvious for the equivalent
form.) Moreover, order intervals are weakly compact. Hence all the struc-
tural conditions of our theorem are satisfied (see Huang and Kreps [8] for
all this).
Our theorem also throws light on a well-known example of Jones. In it
L = L, ([0, 11) is endowed with the weak topology induced by C’( [0, 11).
There are two consumers with utility functions U, (x,) = l; tx, (t) dt,
u2 (x2) = sh (1 - t) x2(t) dt. Initial endowments are o1 = $e, w2 = $e, where
e is a constant positive function. It is immediate that for this economy there
is no equilibrium with continuous prices. Heuristically the failure of
existence in the example is due to L*, not making L* into a vector lattice.
EQUILIBRIA IN VECTOR LATTICES 3
(Recall that L*=C’([O, 11) and L*, = {x~L*:x(t)kO, all t>; since the
supremum of two differentiable functions need not be differentiable, L* is
not a lattice.) The theorem makes this point sharply. Indeed, that L* is a
vector lattice is the only hypothesis of the theorem that Jones’sexample
fails to satisfy. The example has a very pathological feel to it: the obvious
candidate to a quasi-equilibrium price vector fails to qualify simply because
it is decreed not to be continuous. It could be argued that as long as L is
a vector lattice, any problem which is well-posed from the point of view of
existence should have a price space that can be viewed itself as a lattice.
What about the hypothesis that the consumption set, denoted by Xi, is
the positive orthant L, of a vector lattice? An example of Back [6] shows
that it cannot simply be replaced by Xi being closed, convex, and satisfying
A-;+ L, CXiCL,. But we do not know how far it can be relaxed. At any
rate, the hypothesis is central to our proof for the following reason: Let
p,EL*, i=l,..., N be arbitrary and, for any ~30, define n(z) =
sup{~i~Npi.z-i:~iEL+, CitNZI=“,. -’ If L, defines a lattice order then
(because the order has the so-called decomposition property) rc(.) will be
linear. This is essential to us site we construct supporting functionals in
precisely this way.
If a preference relation is continuous and w-uniformly proper in one
topology then it will remain so in any stronger topology. This means that
for a given duality (L, L*) the sharpest theorem lets the topology of the
space be the Mackey topology.
It is easy to generate preferencesto which our theorem is applicable. For
example, let Jc L*, be a(L*, L)-compact with every q EJ o-increasing. Let
u: R x J+ R be jointly continuous with u( ., q) concave and increasing for
every q EJ. Finally, let p be a regular measure on J. Then the function
u(z)= j u(q .z, q) c@(q) defined on the entire L is Mackey continuous,
concave, and o-increasing. The restriction of u to L, yields a preference
relation which satisfies all of our conditions.
Recall from Richard and Zame [ 161 that a sufficient condition for
w-uniform properness on [0, o] is that 2 I be representable by the
restriction to the positive orthant of a continuous, quasiconcave function ui
increasing in the w direction and defined on a convex set which contains
L, and has nonempty interior. Hence, in particular, if L, has nonempty
interior to start with, then w-uniform properness on [0, o] follows from
o-desirability.
Can uniform properness be weakened to pointwise properness (a
necessary condition for individual supportability at any consumption vec-
tor)? An example of Richard and Zame [16] shows that the answer is no.
Nonetheless, recent work by Araujo and Monteiro [4] suggests that in
some particular contexts there may be room for weakening the uniformity
requirement.
b MA.9COLELL AND RICHARD
which contradicts the fact that PEP(~). Hence 7~‘xi= pi. xi and so
rr .z k x .-xi whenever z 2 , xi, as we wanted to prove. Q.E.D.
Denote by A the n - 1 dimensional simplex. We now proceed to specify
a correspondence,
(a) ?ck.z+7r~zforallzE[O,w],
(b) rk ..xf + 71‘xi for all i, and
(cl P E P(-~).
Proof: Part (a). Take any ZE [0, 01. Since [0, o] is compact, so is
[O,z]. Let rc.z= ClaNpi.zi, zi>/O, CleNzj=z. Then 7ck.z3CiaNp:.zi
for all k. Hence lim inf rrk . z 2 CiG N p, zi = rt . z. Since the same inequality
holds for w-z it suffices to show that lim sup rrk. w d II. o. Since
w = CicN .xp this is implied in turn by lim sup rck..$ d n. -Y,. Fix E> 0.
Because every 2 i is a(L, L*)-upper semicontinuous we can assume that
ri + EO>i xf. Hence pf . (?ci+ EW)2 p” xf. Taking limits and letting E+ 0
we get lim sup pf. .xf d pi. xi. But, by Lemma 1, pf. xf = rck x;. Further-
more, pi d rc implies pi. .yi < z . x,. Hence lim sup rck.x: < ?L. x,, as we
wanted.
Part (b). From the definition of rc’, rck. o 3 C,, N pf . x,! for all x’ EX.
Hence lim inf nk . w 2 Cie N pi. xi for all x’ E X. In particular,
lim inf rrk . o 3 z . w. Therefore lim inf C,, N rrk .xF2 C,, N rr . xi. Since
lim sup nk .t$ < rr . .Y, for all i, as proved in part (a), we conclude
7ck. J$ + n xi for all i.
Part (c). For arbitrary E> 0 we eventually have xi + EWti $. There-
fore pk.x,+~p1(.0>pF. .$. Hence pi. x, b lim sup pf . xk. Suppose that
pi.x, > lim sup pf: ..Y; for some i Because xjG N pr .xf 2 xi, ,,,p’. xi we
then have
lim inf 1 pf .*of;> C pi .xi > 1 lim Sup p: . Xf >, lim sup pf . -xf
I 1 I
LEMMA 5. @ is convex-valued.
Proof: Fix an SE A and choose XE X such that ui(xi) =f(s)s, for all i.
We show first that if x’ E X(s) then pi .x; = pi. xi for all ig N and
PEP(X). Indeed, pi.32 pi.xi whenever z 2 i.x,. Since ui(xi)a ui(xj) we
have pi.xid pj.xj for all i. But zieN p, .xib xiEN pi.xi because X’E X.
Therefore pi. xi = pi. xi for all i. But CiG N pi. xi > Cit N pi. x: because
x’ EX. Therefore pi. xi = pi. .x: for i EN. An immediate consequenceof this
is that P(x) = P(x’). From now on we put P(s) = P(x), XE X(s). Clearly,
the set P(s) is convex.
Let t’, t* E a(s) and, correspondingly, take x’, .x2E X(s), p’, p* E P(s).
For any CIE [0, 11, denote p’ = ap’ + (1 - cr)p* E P(s), and rra = ViEN py.
It suffices to show that rcoL(wi - xf) = (car’ + (1 - cl)rc’) . (wi- .x;) =
crt’ + (1 - CI)t*; of course, only the first equality needs proof.
We prove a stronger property, namely that rca.z = (CIII~+ (1 - c1)rc2).Z
for all ZE[O,U]. Because ~‘.~=C~~~pf.xI=C,.~p!.xf and ?r*.a=
c,,,pf.xf we have (~1~‘+(1--)7~*).0=~~,~p~..~~=.~.w. The last
equality follows from pz E P(s). For ZE [O,w] and C(E[0, l] let ~9, igN,
be such that 2’ > 0, Cit N ~4 = L, and rca. z = Cic N p’. ~-4. (Note that the
superscript in the definition of rr’ is realized because [0, z] c [0, o] is
a(L, L*)-compact.) Then
REFERENCES